This paper proposes a statistical method (correlation analysis) to compare the similarity between two waveforms. In this paper, we try to detect an arcing fault using the correlation coefficient between the current signature measured and a database of arcing current signatures. In this article, we present the database of current signatures which contains arcing signatures for different types of loads (resistive, vacuum cleaner, dimmer, switching power supply). Then, the correlation is done on the measured current considering a half-period or a period of the signal. The correlation coefficient gives a value between-1 (perfect negative correlation), 0 (no correlation) and +1 (perfect positive correlation). For the first method, an arcing fault is detected if the value is close to +1. The second proposed method is based on the correlation between two consecutive periods of the signal. An arc fault is detected if the value is different to +1. The results show that it is possible to detect an arcing fault using the correlation coefficient with different performances for different types of load.